Figures.Rmd

---
title: "A Pass at Figures"
author: "Aaron Meyer"
date: "June 28, 2016"
output: 
  html_document:
    keep_md: true 
---

```{r setup, include=FALSE, echo=FALSE}
library(dplyr)
library(ggplot2)

MB231 <- read.csv("./data/MB231.csv", stringsAsFactors = FALSE) %>% tidyr::separate(X, into = c("matrix", "drug"), sep = " ")
SKBR3 <- read.csv("./data/SKBR3.csv", stringsAsFactors = FALSE) %>% tidyr::separate(X, into = c("matrix", "drug"), sep = " ")

MB231$matrix <- as.factor(MB231$matrix)
MB231$drug <- as.factor(MB231$drug)
```

**Note that text describes the figure following it.**

## Sample information

### Matrix
- TCPS:	tissue culture plastic
- gel:	stiff 2D PEG-PC gel
- single:	single cells in 3D PEG gel
- sph: spheroids in 3D PEG gel
- serum: spheroids in 3D PEG gel plus serum

### Drugs
- DMSO	
- lap:	lapatinib
- tems:	temsirolimus
- soraf:	sorafenib

2 biological replicates, each with 2 technical replicates	
DMSO control can be used for all 3 drugs (equalized DMSO concentration across all)	

This plots the drug response in each matrix condition separately, against pMek. From this it's possible to see a relationship between viability and pMek that exists across matrix conditions.

```{r mek plot matrix}
ggplot(MB231, aes(x = MEK1, y = prolifTwo, color = drug, shape = matrix)) + 
  geom_point() + facet_wrap( ~ matrix) + 
  coord_trans(x = "sqrt", y = "sqrt") + theme_bw() + ylab("Viability (RU)") + xlab("pMek")
```

A linear regression model clearly shows that pMek significantly explains some of the variance in viability. Note that the model explains about 30% of the variance, so there certainly are changes in drug response not completely captured by the model still.

```{r lm model, fig.width=4}
# Scaling the phospho-measurements by z-score
MB231d <- dplyr::select(MB231, prolifTwo, CREB, EGFR, NFkB, p38, AKT, JNK, MEK1, ERK1.2) %>%
  scale(center = TRUE, scale = TRUE) %>% data.frame

# Applying the linear model
model <- lm(prolifTwo ~ . + 0, data = MB231d)

# Summarize the model
model.sum <- (summary(model))

# Print the model summary
print(model.sum)
```

Plot of MB231 model parameters.

```{r, fig.width=4}
# Extract the model coefficients for plotting
plotD <- data.frame(model.sum$coefficients)

ggplot(data = plotD, aes(x = Estimate, y = log10(Pr...t..), label = rownames(plotD))) + 
  geom_point() + theme_bw() + geom_text(nudge_y = 0.06) + scale_y_reverse() + xlim(-1.2, 1.2) +
  ylab("Log10 P-Value") + ggtitle("MB231 Viability Linear Model") + 
  geom_errorbarh(aes(xmin = Estimate - Std..Error, xmax = Estimate + Std..Error))
```


# SKBR3

I've plotted the SKBR3 data with respect to pMek just for consistency with the MB231 data. There are considerable matrix-dependent differences in drug response, but they aren't really explained by pMek.

```{r mek plot matrix skbr}
ggplot(SKBR3, aes(x = MEK1, y = prolifTwo, color = drug, shape = matrix)) + geom_point() + facet_wrap( ~ matrix) + 
  coord_trans(x = "identity", y = "identity") + theme_bw() + ylab("Viability (RU)") + xlab("pMek")
```

A linear model with the SKBR3 measurements doesn't really pick up much.

```{r, fig.width=4}
# Scaling the phospho-measurements by z-score
SKBR3d <- dplyr::select(SKBR3, prolifTwo, CREB, EGFR, NFkB, p38, AKT, JNK, MEK1, ERK1.2, STAT5, p70s6k) %>%
  scale(center = TRUE, scale = TRUE) %>% data.frame

# Applying the linear model
model <- lm(prolifTwo ~ . + 0, data = SKBR3d)

# Summarize the model
model.sum <- (summary(model))

# Print the model summary
print(model.sum)
```

Plot of SKBR3 model parameters.

```{r, fig.width=4}
# Extract the model coefficients for plotting
plotD <- data.frame(model.sum$coefficients)

ggplot(data = plotD, aes(x = Estimate, y = log10(Pr...t..), label = rownames(plotD))) + 
  geom_point() + theme_bw() + geom_text(nudge_y = 0.03) + scale_y_reverse() +
  ylab("Log10 P-Value") + ggtitle("SKBR3 Viability Linear Model") + 
  geom_errorbarh(aes(xmin = Estimate - Std..Error, xmax = Estimate + Std..Error))
```